‹ All projects

Real-time vision

Real-Time Hand Gesture Recognition with MediaPipe and Machine Learning

An end-to-end, real-time gesture recognition system for touch-free human-computer interaction in complex scenes.

Period
Oct 2025 – Nov 2026
Stack
  • Python
  • MediaPipe
  • OpenCV
  • scikit-learn
01 / 04Problem
  1. 01 · Problem

    Touch-free interaction.

    An end-to-end, real-time gesture recognition system built for touch-free human-computer interaction in complex scenes.

  2. 02 · Features

    21 landmarks, normalised.

    MediaPipe extracts hand landmarks from each frame. They are normalised for position and scale, so the classifier sees the shape of the hand rather than where it is or how large it appears.

  3. 03 · Models

    Three classifiers, one written from scratch.

    Logistic Regression, a Decision Tree and a k-nearest-neighbours classifier written from scratch were compared.

  4. 04 · Robustness

    Built for poor lighting.

    Adaptive gamma correction and median filtering improve hand detection in low-light and noisy conditions. The pipeline reaches 93.8% end-to-end test accuracy.

Key result

93.8%

End-to-end test accuracy.